Appendix 5: Effect Sizes
To illustrate the impact of different factors on attainment or social behaviour in Year 1 effect sizes
(ES) were calculated. Effect sizes are most commonly used in experimental studies and
essentially measure the strength of mean differences. Glass et al., (1981) define ES as:
ES = (mean of experimental group)-(mean of control group)/pooled standard deviation
Or ∆ = XExp - XCont
SDpooled
Effect sizes were calculated for different child outcomes, using both the child level variance and
coefficients for predictors included in the multilevel statistical models adopting the formulae
outlined by Tymms et al., (1997).
For categorical predictors (e.g. gender or ethnicity) the effect size was calculated as:
ES = categorical predictor variable coefficient / √child level variance
Or
Δ = β1
σe
For continuous predictor variables (e.g. child age in months), the effect size describes the
change on the outcome measure produced by a change of +/-one standard deviation on the
continuous predictor variable, standardised by the within school SD, adjusted for covariates in
the model - the level 1 SD:
Δ = 2 β1*SDx1 where x1=continuous predictor variable
σe
Effect sizes can be useful for comparisons between different studies but interpretations must be
made with caution and with reference to the outcomes concerned and controls used in models
(Elliot & Sammons, 2003). For further discussion of effect sizes see Coe (2002). Effect sizes for
some categorical measures in the EPPE research are large but apply to small numbers of
children (e.g. the very low birth weight group or specific ethnic groups).
48